mirror of
https://github.com/vllm-project/vllm.git
synced 2025-10-20 14:53:52 +08:00
[ROCM][AMD][TRITON] Halving warps number for fw_prefill to reduce spilling (#12713)
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com> Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
This commit is contained in:
committed by
GitHub
parent
b3a0d01e45
commit
64862d106e
@ -11,7 +11,7 @@ from vllm.platforms import current_platform
|
||||
|
||||
# Static kernels parameters
|
||||
BASE_BLOCK = 128 if current_platform.has_device_capability(80) else 64
|
||||
NUM_WARPS = 8
|
||||
NUM_WARPS = 4 if current_platform.is_rocm() else 8
|
||||
|
||||
# To check compatibility
|
||||
IS_TURING = current_platform.get_device_capability() == (7, 5)
|
||||
|
Reference in New Issue
Block a user